AI Models
Grok is great as a frontier-level model, but Composer models offer an incredible intelligence-per-dollar ratio.
A lot of people don’t need the absolute frontier. They just need something fast, cheap, and ~90% as capable as the frontier models, which is more than enough for everyday coding.
Composer is exactly that. Composer 2.5 is just starting to feel a bit outdated now. So please release Composer 3 soon.
I aproove !
Composer hits a great sweet spot between intelligence, speed, and cost. A newer version that closes the gap with frontier models while keeping that efficiency would be extremely useful for everyday coding.
It looks like Grok 4.7 is next in the pipeline.
i am looking how will the composer 3 will work?
also will grok 4.7 release first or composer 3?
+1
Since the policy seems to be that no model will be added that’s cheaper per-token than the current Composer model, then we need Composer 3 in order to compete with GLM-5.3 Flash (and to a lesser extent, Deepseek v4 Flash 0731).
Otherwise, we’ll have to move over to Codex in order to get the GPT stack + Astra + GLM-5.3 Flash once the OpenAI departure happens in November - keeping in mind we’ll also be losing Luna at that time.
Love Composer 2.5 and would love to see Composer 3.0 soon. Please can we have some updates on any plans, yes or a no, will there even be a Composer 3.0?
Composer 2.5 is a phenomenal model. I use it as my daily driver, with Opus 5 for deep work.
Definitely expect the Composer models to remain a Cursor staple.
I tried Grok 4.6, it can’t even express itself, the phrasing is so artificial and awkward.
I will undoubtedly switch from Cursor to something else if Composer gets deprecated.
There’s a chance that rather than Composer 3 we get a “Grok Lite” or something of that nature - basically a Luna equivalent in the Grok line. After all, the work that went into the original Composer 3 seems to have gone into Grok 4.5.
That said, I fully agree with you that I will be doing the same if we don’t get a competitive model in that market space. There’s a reason why the GLM-5.3 Flash and Deepseek v4 Flash 0731 request threads are as popular as they are at the moment, and Cursor staff seems to be actively avoiding replying in any of these relevant threads unless someone calls them out for not replying.
Deepseek thread still doesn’t have a reply, and the GLM Flash thread has a “non-reply” reply (it didn’t acknowledge GLM-5.3 Flash, it acknowledged the GLM-5.3 base model despite the thread being specifically about Flash and there already being a thread about GLM-5.3 base) - and it only came after they were directly called out. Less popular and more recent threads than those and this one have gotten replies at least acknowledging the topic of the thread, but the four major and upvoted model request threads (Deepseek v4 Flash, GLM-5.3 base + Flash, and Composer) don’t really have anything.
All of that said, I don’t blame the community management staff for the lack of communication - they’re probably getting screwed by upper management in terms of what they can say in these threads.
EDIT: The Deepseek thread apparently had a reply, but like the other one, it’s a non-reply reply and no follow-up despite how long it’s been since that thread’s been up.
I found this thread while searching for Composer 3 given how pleased I am with 2.5 and how it hyper accelerated my work. Only to see there might not be a 3.0.
One of the things I love about Cursor is the wide range of models you can plug and play.
If Cursor becomes nothing more than a Grok harness, see you later alligator.
Tried Grok 4.6 Extra High, useless and nonsensical. Gave Opus 5 Max the same task and hit a grand slam. It’s not even funny.
Rather than “see you later, alligator”, I think you mean “in a while, Grokodile” 
I so agree with this. Right now only 5.6 Luna competes at the same price point and maybe Gemini 3.8 Flash. OpenAI models are going away as we all know. And top of that for those of us on Enterprise plans , the value proposition of cheaper third party models is completely demolished due to the 25 cents markup.
It should be a no brainer to release a Composer 3. Or atleast a Grok Lite/Flash.
Composer 2.5 is more reliable for coding than Grok 4.6. I can have Grok Audit but no code. He makes great images, but his coding is full of mistakes and half done. If I have Sonnet or Opus do work, Composer audits and fixes bugs, not Grok.
Cursor was a developer tool for the community that everyone could use, now its xAI and I will be straight up. I have 4 Ultra accounts and I have already started looking at other platforms because the tool is what I use to code, its not a pair of Nikes or Makita power tool. Its a work bench and if my work bench loses access to models and Composer too, there is literally no reason to use a tool that has been nerfed due to branding. I would guess there are many power users like me that feel the same.
Just the name alone gives me confidence.
Zero replies from the staff on this seems a bit worrying. If Cursor is going to devolve into a Grok harness, please let me know so I can pack my bags early.
Grok and Composer are two different model weight classes, and we’re excited to support both sizes and weights. Composer 2.5 remains supported and it is still our plan to release new models of this size going forward!
Composer 3 ought to reveal some information; that would give us some confidence.
@Colin thanks for your reply.
Composer 2.5 remains supported and it is still our plan to release new models of this size going forward!
That’s good to know. It would be nice if we could have a bit further info on that, i.e. if development continues on the Composer 3 model or it has moved to something that will be similar to some kind of “Grok Lite” in the future.
Cursor was a developer tool for the community that everyone could use, now its xAI and I will be straight up. I have 4 Ultra accounts and I have already started looking at other platforms because the tool is what I use to code, its not a pair of Nikes or Makita power tool. Its a work bench and if my work bench loses access to models and Composer too, there is literally no reason to use a tool that has been nerfed due to branding. I would guess there are many power users like me that feel the same.
@Pawsnfound I have one Pro+ account and feel exactly the same.
I’ve been waiting for Composer 3, Composer 2.5 is fantastic.